A computer model called a storm’s landfall days before most of its rivals. The people who use it are still not sure why it works.
When Hurricane Isaias formed in early October, Google DeepMind’s model pointed to the western Florida Panhandle. Conventional models leaned toward southeastern Louisiana, NPR reported.
The National Hurricane Center took notice. CNN reported that the center based its early forecast closely on the AI model.
The center forecast Isaias as a Category 2 hurricane from its first advisory. Many physics-based models showed a weaker storm on a different track, CNN reported.
Why Miami Is Watching Closely
The National Hurricane Center works from Miami-Dade County. Its forecasts are the ones millions of people depend on.
Michael Lowry, a Miami meteorologist who has advised FEMA and the center, calls DeepMind the best-performing computer model forecasters have seen.
The center first used it in official forecasts last year. Director Michael Brennan says forecasters blend AI with traditional models.
“I think the most important thing is to make sure that the models are trained on the most accurate data we have,” Brennan told CNN.
A track is where a storm goes. Intensity is how strong it is when it gets there. Forecasters need both, and they are different problems.
Earlier AI models did well on track and poorly on intensity, CNN reported. The center worked with Google to train DeepMind on its own best-track analysis.
Brennan says the AI models have been more consistent from run to run. Steady forecasts are easier to trust and easier to communicate.
Miami’s Local10 told viewers in September the model was once again leading the pack in 2026.
The model is not perfect. It nailed Isaias’s track but struggled with intensity, NPR reported. The center’s blended forecast did better on storm strength.
The Black Box Problem
DeepMind learns from thousands of past storms. It does not solve the physics equations that older models use.
That makes its reasoning hard to inspect. Lowry calls it a black box.
Brian McNoldy, a hurricane researcher at the University of Miami, says forecasters feel uneasy. They are relying on a system they do not understand, and millions depend on the result.
He also called the model’s debut impressive. New models usually take years to refine, and this one “came out of the gate in first place.”
Only two models have shown steady accuracy, Lowry said. They are DeepMind and the European Centre for Medium-Range Weather Forecasts.
The pace is the real shift. Traditional models are updated once a year at best. DeepMind improves in months or weeks.
Lowry says intensity errors have fallen to about half those of the Hurricane Katrina era.
He does not fear replacement. “If anything, I think it is a boon for our business,” he said.
A Local Test of the Same Question
The University of Miami has tested this trade-off on its own. MAIN covered a University of Miami study of AI hurricane forecasting.
Rosenstiel School professor Sharan Majumdar compared the European center’s AI system with its traditional model. The AI did better days ahead, at spotting storms.
Closer to formation, it did worse. Majumdar framed AI as a complement, not a replacement.
The study covered the 2024 season, when 18 tropical cyclones developed. It looked at the waves that often organize into hurricanes.
Isaias fits that pattern. The AI saw the track early, and the blended forecast handled strength better.
Miami has a stake in the answer. A forecast that is right days out gives hospitals, ports and condo boards time to act.
Forecasters are not choosing between AI and physics. They are learning where each one fails.
Sources:
- NPR: AI is changing hurricane forecasting; National Hurricane Center blends DeepMind with physics models (Oct. 2026)
- CNN via KSL: Why the National Hurricane Center bet big on AI to forecast Isaias (Andrew Freedman, Oct. 8, 2026)
- National Hurricane Center: Location
- MAIN: UM Study: AI Could Improve Hurricane Forecasting, With Limits (Aug. 28, 2026)
